fig1

Figure 1. The input image from the KITTI dataset (top); the baseline MonoDepth2^[22] (M, ResNet50, without pre-training) depth prediction; (middle) and our result (bottom).